xTweetDocsYour project’s reply game,
running on its own
xTweet watches the accounts that matter to you, writes replies in your project’s voice and sends them from your X account, at the pace and inside the limits you set.
no agency retainer · no content calendar · you keep the account
per second only works if you can evict fast, and nobody has solved eviction
token pricing hides the 40% of gpu time that is just weights sitting there
every provider says this right up until they see their own utilization chart
Four steps, then it is off your desk
Setup takes an evening. After that you are reading a feed, not writing one.
Connect your X
Link the account the replies go out from. Nothing is posted until you finish the setup.
Point it at sources
Add the accounts worth replying to, one by one, as a list or by importing an X list. Group them into folders, block the ones you never want touched.
Set the rules
Reply rate per hour, gap between replies, cap per author, quiet hours in your timezone. You decide how loud it gets.
It runs
New posts get read, drafted and answered in your voice, inside the limits you set. You watch the feed, not the clock.
Built so the replies do not read as automated
Everything that can be checked by code is checked by code, not asked for in a prompt.
It learns how you write
Start from a preset or describe the voice in your own words. Edit a reply and the edit becomes a reference the model writes toward, so the account keeps sounding like the people behind it, not like a tool.
Human pacing, not a firehose
Quiet hours, a minimum gap between replies, a cap per author and a ceiling per hour. All in your timezone, daylight saving included, so nothing goes out at 4am local and nothing spams one thread.
Eleven angles
Push back, add a number, tease, ask what comes next. The angle is picked per post, so the account does not repeat one move all day.
Filters before it ships
Stock phrases, hashtags, links, restating the original post, a trailing period. Cut by code before the reply exists, not asked for in a prompt.
Reads images and quotes
Screenshots, charts and quoted posts go into the context. Numbers, names and brands read off an image never make it into the reply.
Short answers
Whose account do the replies come from?
Yours. You connect the project account once, and every reply goes out from it, inside the limits you set.
Will it sound like a bot?
Stock openers like "here is why" and "game-changer", hashtags, links, a trailing period and anything that just restates the original post are cut by a check before the reply is used.
How much control do we get?
Reply rate, gap between replies, cap per author, quiet hours, which accounts are in scope, which are blocked, which angles are allowed and what the voice sounds like. Every one of those is a setting, not a support ticket.
Who is this for?
Crypto and AI teams whose presence on X matters and whose founders do not have four hours a day to spend in replies.